Hoeffding Tree Method with Feature Selection for Forecasting Daily Demand Orders

In supply companies, the task of forecasting daily demand orders is essential in planning and provision processes to achieve the consumers’ requests on time and reduce the costs. Machine learning (ML) methods play an important role in developing effective tools for forecasting problems in many fields. Previous works of forecasting daily demand orders have proposed to use a set of effective ML algorithms and methods. However, there is still room for improving the accuracy result using more effective computational methods. In this paper, an effective approach that uses a Hoeffding tree method with an information gain ratio feature selection algorithm is proposed to predict the orders of daily demand products in an interval period of time. The approach is assessed on a real public dataset of the Brazilian logistics company that is gathered through 60 days. After selecting the most important features, the Hoeffding tree method is trained and tested on this dataset using a 10-fold cross-validation technique. The experimental results show that the proposed approach is able to predict the daily demand orders and achieves a competitive accuracy result with a small number of features compared to the recent related works.